Transforming data-driven teams into engines of business value. Head of AI,Data,Analytics @YouTube (ex-Airbnb,eBay)20+ years in tech. Ruthless Curiosity ๐ฎ๐ฑ๐บ๐ธ
@cquilboss what i keep seeing is one team running agents while the team next door still pastes into ChatGPT. do you measure a company by its best team or its slowest?
@krishramineni honest question: when you hire now, do you look for people who can spot bad AI work fast? that feels like the skill that matters most right now
@joshzemon Small model + good harness, ok. But who owns the eval set?
If the harness gates on "tests pass" and not on a product metric with a kill switch behind it, you just ship cheaper shit show
the denials log is where i'd push. append-only is fine but an actual person has to actually read it every week or it's just becoming row in the db
also every agent in the registry needs a kill path the owner has tested before it gets write access. most teams find out they don't have one in the middle of an omg incident